F5F Stay Refreshed Hardware Desktop Referring to reducing clock speed for faster response times rather than focusing on megahertz values.

Referring to reducing clock speed for faster response times rather than focusing on megahertz values.

Referring to reducing clock speed for faster response times rather than focusing on megahertz values.

Pages (2): 1 2 Next
B
Bloody_TearS
Member
180
11-18-2018, 01:37 PM
#1
I've been exploring RAM options for a new build. What caught my attention was that prices don't seem to jump much with higher speeds. In theory, you could find kits around 6400 CL30 for as low as CL26. I'm considering a different approach—using top-tier memory like 8000 CL38 and lowering it to 6400 to focus on tighter clock speeds instead of chasing minor speed gains. Would running it at 2600MHz below its rated frequency yield noticeable improvements, perhaps with something like 6400 CL22? Or would the outcome be similar or worse? Also, how complex would this process be compared to regular memory overclocking? My main concern is diving into the technical details of secondary and tertiary timings; I'd rather avoid that unless absolutely necessary. I'm comfortable adjusting primary timings until it fails or crashes, but I don't want to get tangled in secondary settings. I prefer sticking to the straightforward method: reduce primary timings until it stops working or fails Memtest86, then fine-tune with a safety margin. If the results don't meet expectations or prove too risky, I'll settle for the best available 6400 CL30 option. And regarding sourcing better memory dies, let me know—I've been out of the loop for a while. Thanks.
B
Bloody_TearS
11-18-2018, 01:37 PM #1

I've been exploring RAM options for a new build. What caught my attention was that prices don't seem to jump much with higher speeds. In theory, you could find kits around 6400 CL30 for as low as CL26. I'm considering a different approach—using top-tier memory like 8000 CL38 and lowering it to 6400 to focus on tighter clock speeds instead of chasing minor speed gains. Would running it at 2600MHz below its rated frequency yield noticeable improvements, perhaps with something like 6400 CL22? Or would the outcome be similar or worse? Also, how complex would this process be compared to regular memory overclocking? My main concern is diving into the technical details of secondary and tertiary timings; I'd rather avoid that unless absolutely necessary. I'm comfortable adjusting primary timings until it fails or crashes, but I don't want to get tangled in secondary settings. I prefer sticking to the straightforward method: reduce primary timings until it stops working or fails Memtest86, then fine-tune with a safety margin. If the results don't meet expectations or prove too risky, I'll settle for the best available 6400 CL30 option. And regarding sourcing better memory dies, let me know—I've been out of the loop for a while. Thanks.

S
61
11-18-2018, 01:37 PM
#2
latency depends on frequency. For example, 8000 CL38 equals 9.5ns and 6400 CL30 equals 9.4ns. Looks like they match.
S
Spiegelmeister
11-18-2018, 01:37 PM #2

latency depends on frequency. For example, 8000 CL38 equals 9.5ns and 6400 CL30 equals 9.4ns. Looks like they match.

K
KizuPvP
Member
137
11-18-2018, 01:37 PM
#3
For AM5, exceeding 6000Mhz memory tends to cause instability, making it likely the system won’t start. The top recommendation is to purchase 6000 CL30 modules.
K
KizuPvP
11-18-2018, 01:37 PM #3

For AM5, exceeding 6000Mhz memory tends to cause instability, making it likely the system won’t start. The top recommendation is to purchase 6000 CL30 modules.

A
agarmor
Member
223
11-18-2018, 01:37 PM
#4
You could try it, but I wouldn't expect it to give you drastically lower latency than if you just tried tightening timings on a DDR5-6400 kit. The Silicon Lottery is called a lottery for a reason. No one can tell you if a particular RAM kit is going to be a winner. And you also would need to have a solid IO die on your 9950X3D to facilitate this as well, making your odds of achieving ridiculously low timings even lower. And at the end of the day, you're not going to see as much benefit as someone with a standard CPU, because of the 3D V-Cache. The whole point of that cache is to give the CPU low-latency access to the most needed data without needing to go out to the RAM. RAM speed and latency doesn't scale as much when you have a 3D V-Cache chip. This was shown way back with the 5800X3D. You can get away with slower, higher latency RAM and it won't compromise the performance because the cache is circumventing the RAM bottleneck. If you want to do this just for fun, then have fun, but there's no practical reason to do this beyond OC for the sake of OC.
A
agarmor
11-18-2018, 01:37 PM #4

You could try it, but I wouldn't expect it to give you drastically lower latency than if you just tried tightening timings on a DDR5-6400 kit. The Silicon Lottery is called a lottery for a reason. No one can tell you if a particular RAM kit is going to be a winner. And you also would need to have a solid IO die on your 9950X3D to facilitate this as well, making your odds of achieving ridiculously low timings even lower. And at the end of the day, you're not going to see as much benefit as someone with a standard CPU, because of the 3D V-Cache. The whole point of that cache is to give the CPU low-latency access to the most needed data without needing to go out to the RAM. RAM speed and latency doesn't scale as much when you have a 3D V-Cache chip. This was shown way back with the 5800X3D. You can get away with slower, higher latency RAM and it won't compromise the performance because the cache is circumventing the RAM bottleneck. If you want to do this just for fun, then have fun, but there's no practical reason to do this beyond OC for the sake of OC.

I
IronMaster720
Junior Member
1
11-18-2018, 01:37 PM
#5
Fair enough, I understand it doesn’t really matter much, but I’m just trying my best. That claim about AM5 not handling speeds over 6000MHz is something I’ve never heard before—I see a lot of people using 6400 and enthusiasts pushing to 8000. I thought it was mainly about choosing a solid motherboard with memory overclocking capabilities, which will be addressed later. Yeah, I’m not planning to do much overclocking at all. Modern computers come out pretty well ready, so I see overclocking more as a way to cause issues and wear out parts faster. For the CPU and GPU, I plan to keep things simple and maybe just let them run normally.
I
IronMaster720
11-18-2018, 01:37 PM #5

Fair enough, I understand it doesn’t really matter much, but I’m just trying my best. That claim about AM5 not handling speeds over 6000MHz is something I’ve never heard before—I see a lot of people using 6400 and enthusiasts pushing to 8000. I thought it was mainly about choosing a solid motherboard with memory overclocking capabilities, which will be addressed later. Yeah, I’m not planning to do much overclocking at all. Modern computers come out pretty well ready, so I see overclocking more as a way to cause issues and wear out parts faster. For the CPU and GPU, I plan to keep things simple and maybe just let them run normally.

A
apilot0404
Member
178
11-18-2018, 01:37 PM
#6
It was significantly worse for the 7000 series, yet it remained somewhat unpredictable compared to the 9000 series. A solid motherboard helps boost your chances, but it doesn't ensure success. If you're willing to take the risk, I'd suggest opting for a verified kit to increase your odds.
A
apilot0404
11-18-2018, 01:37 PM #6

It was significantly worse for the 7000 series, yet it remained somewhat unpredictable compared to the 9000 series. A solid motherboard helps boost your chances, but it doesn't ensure success. If you're willing to take the risk, I'd suggest opting for a verified kit to increase your odds.

S
Sihere
Member
187
11-18-2018, 01:37 PM
#7
Most benchmarks show 6400 leads the pack. It consistently outperforms 8000 across the board, though the differences are usually small. If I purchase two 32x2 sticks of 6000 CL26 for a combined 128GB, that would be ideal. G.Skill Neo, based on my experience, I’m confident in this choice. I’d opt for 6400 if I had 32GB drives, but currently 24GB seems to be the highest available above 6000. Yes, it’s a huge amount of RAM—my workflow involves heavy tasks like Unity, Blender, and lots of encoding, compiling, and compressing files, which gives me my justification.
S
Sihere
11-18-2018, 01:37 PM #7

Most benchmarks show 6400 leads the pack. It consistently outperforms 8000 across the board, though the differences are usually small. If I purchase two 32x2 sticks of 6000 CL26 for a combined 128GB, that would be ideal. G.Skill Neo, based on my experience, I’m confident in this choice. I’d opt for 6400 if I had 32GB drives, but currently 24GB seems to be the highest available above 6000. Yes, it’s a huge amount of RAM—my workflow involves heavy tasks like Unity, Blender, and lots of encoding, compiling, and compressing files, which gives me my justification.

A
53
11-18-2018, 01:37 PM
#8
Some reports indicate performance drops when using 4 sticks on AM5. For 128GB RAM, consider a 2x64GB configuration.
A
alexsandar_123
11-18-2018, 01:37 PM #8

Some reports indicate performance drops when using 4 sticks on AM5. For 128GB RAM, consider a 2x64GB configuration.

N
Nienke_2002
Senior Member
621
11-18-2018, 01:37 PM
#9
Understanding that using four memory modules can affect stability, especially with DDR5-6000, is important. DRR5 tends to struggle with more than two sticks, which is why the officially recommended speed for four sticks on the 9950X3D is DDR5-3600. You might find success with DDR5-4800, 5200, or 5600, but using four sticks is uncommon and may not work reliably.
N
Nienke_2002
11-18-2018, 01:37 PM #9

Understanding that using four memory modules can affect stability, especially with DDR5-6000, is important. DRR5 tends to struggle with more than two sticks, which is why the officially recommended speed for four sticks on the 9950X3D is DDR5-3600. You might find success with DDR5-4800, 5200, or 5600, but using four sticks is uncommon and may not work reliably.

R
redstonersven
Member
210
11-18-2018, 01:37 PM
#10
R
redstonersven
11-18-2018, 01:37 PM #10

Pages (2): 1 2 Next